Speed increases with modern processors

I am extremely curious to know what degree of a speedup (or not) I could expect were I to upgrade to a more modern computer.

I currently have an old dual Athlon 1900+ MP system (two athlon32 1.6Ghz processors, SMP).

I had always previously held myself to the rule of only upgrading when I could get a new system that was at least 4 times as fast as my current one.

I am considering getting a Athlon64 X2 4200+ (2.2Ghz dual core) rig, but I am tearing my hair out trying to determine if it would really be worth the money or only marginally faster.

I have had my current system for over 3 years, so you would think that a modern system would be quite a bit faster… but the Ghz speed difference is minor. I know the newer athlon64s do more per clock, but the question is how much… If my new system only ends up being less than twice as fast, it wouldn’t be worth the money.

Does anyone have any thoughts or real world experienience in the magnitude of rendering speedup I could expect were I to take the plunge and upgrade?

I would really appreciate your comments on this matter. Thanks in advance! :wink:

I don’t know what the actual speed increase, but I think it would be very much worth it. Upgrade!

you will see a big increase because the cache (512kb per core) and the on die memory controller.
however, around here, the 4200+ is $450. but the 3800+ is $350. it is a 200mhz difference. buy a 3800+, save $100 and bump it up 200mhz. there is a 99.9% chance the cpu will be able to do it.

that said,
you are wanting to buy a 4200+, they are pricey. amd is releasing socket am2 in a few months, and will be faster. i would wait, buy an am2 motherboard, and buy a 4400+ that has 1mb of cache per core.

My main desktop is an Athlon XP 2200+, and I recently put together an A64 3200+ (the slowest A64 IIRC) system for my HTPC, since I didn’t want to be stuck with an old architecture but was still cost-concious. It is running in 32-bit mode, and the speed increase is very very noticable over the XP. I can’t give any real benchmarks, but the feel is something like 2x as fast. Emerges in gentoo are suprisingly faster, maybe 3x faster for some things. The A64 memory system is very very good. The scaling for SMP on A64 is the best that has ever been available in the off-the-shelf consumer arena. With linux’s NUMA support, a dual dual-core system would kick serious arse… <droool> I am giong to upgrade my desktop to A64 as soon as I can afford to.

Your hardware is import but you can get by with less PC power if you…

Apply a good basic 3 point lighting setup. Use layer based object lighting. Try using buffered shadow lamps instead of raytraced shadows. You can still use raytraced reflections and a moderate AO setup. You will see your render speeds increase while you get excellent renders.

Yafray GI lighting, high detail textures and high poly scenes will bring even the mightiest PC hardware to it’s knees.

Blend on!

Thanks for the comments. I really appreciate any advice!

From what research I’ve been able to find (looking on tomshardware.com and the like)… looking at the benchmarks, it does seem that the memory subsystem is up to 12-15 times faster than my old DDR133 registered (my tiger SMP board requires registered memory (more reliable, but slower than unregistered) ram.

In upgrading, especially if I spend a bit more and get higher-grade ram (CAS 2, etc) I might expect to see a very significant improvement.

However, computational speed looks to only improve 2-3 times, (which isn’t all that great for 3+ years of moore’s law!!!) …of course, I can’t really tell, because I’m comparing apples to oranges and then oranges to kiwis, so to speak, because there are no comparative benchmark charts I can find anywhere comparing and old athlon 1900+ with modern hardware. (I did find a comparison between athlon 1900 and athlon 3200, and then one with 3200 vs. 4200X2, but they were using different benchmark suites)… so I’m ultimately guessing whether or not I’d only see a 2x speedup or whether it would be more worth it.

If only I had unlimited funds, this wouldn’t be an issue for me, but I try to only upgrade my system every few years.

As far as waiting for the new AMD M2 chipset, it might not be entirely optimal when it first comes out (usually best to wait for a few revisions of motherboards to really take advantage of a new interface)… and I’ve heard a lot of talk about how DDR2 memory (which the M2 athlons will use) is actually SLOWER than current DDR (although it’s supposed to eventually scale to higher bandwidth, it has considerably higher latency… not good for applications tending more towards random access, and that don’t predictably consume large chunks of memory (which I suspect is the case with blender)).

As far as which dualcore to buy, I’ve also heard that 200mhz faster is better than twice the cache, but maybe this is untrue.

Oh, also I can get a x2 3800 for $300 and an x2 4200 for $350. %10 faster is probably worth $50 cause I’m not likely to be upgrading again for at least another couple years.

try to get a 4400+ bc they have 1mb of cache per core vs 512k of anything less.

moores law states that circuit density increases, not speed.

spending more on ram isnt really worth it. it might shave off 10-20s of a 20min render. its the difference between 5500mbps and 6000mbps. not really too important. i have ocz platinum 2-2-2-5 ram in my comp right now, but its designed for oc not low timing. i shelled out 250 for these and i could have got the same in value ram for 100.

if you can afford it, get a dual core opteron and oc it. they will do you nicely, or even stock. they are more reliable.

i have a 3200+ system, just waiting for a vid card, i bought it for oc, not just for stock. people have reached 3ghz on them with water. the slowest 939 is a 3000+ at 1.8ghz.

again, id wait for am2, because even if you dont buy it, the prices for 939 will drop ~20%. cache will probably do you better than speed in blender.

i know the crunch your in, i upgrade every few years as well and dont have that much to throw into it.
i tihkn i do well though, my $1200 should compete with a $3000 setup once its complete.

ddr2 is slower but its faster. it runs ~4-4-4-12 for timings but it does this at ~800mhz, much faster than any ddr memory.

I had a 1.3 Ghz Thunderbird with 256MB ram. Then I broke the motherboard. My dad got a new one that made my processor speed 1100Mhz (bus speed is slower) and my ram speed slower, from 233Mhz to 133Mhz. But I got an upgrade to 512MB ram, and my rendering speed increased dramaticly.

the difference between 256 and 512 is much greater than 1Gb to 2Gb.


I take the same view about not upgrading my machine until absolutely necessary or if the upgrade is cheap. I have done some extensive research into the matter - that means I’ve looked at one or two benchmarks sites :slight_smile: - and I’ve come to the same conclusion that the modern processors are about 3-5 times faster than what you and I have.

I personally don’t consider that worth an upgrade. I have to say though, that the price seems to vary linearly with speed. A machine 5x faster than mine is about 5x more expensive.

I have a 1.25GHz machine with 1GB Ram and I’m pretty content with it. I have a low end Radeon 9200 so I have to keep my models low res but I do that anyway.

I think the best thing you can do is to use optimized software to get the biggest increase in performance. I’ve been trying to get 3delight (Renderman renderer) set up to use with Blender and most of it is working fine and it can render much more efficiently than Blender. It can render millions of polys while using around 15MB Ram in the same time Blender takes to do a few thousand polys with 60MB Ram. Blender would use nearly all my Ram doing a million polys.

I’d say stick with the machine you have and possibly get a cheap graphics card upgrade if you feel yours is too slow. Otherwise I don’t see the point. I reckon you’d be far better using some money to use a fast computer cluster like respower or similar. That way you get a significant render time speed up while not burning your processor.

Check out Tom’s Hardware guide’s Benchmarking Marathon, in which they side-by-side test everything from a 100Mhz Pentium to an Athlon XP 3000+ with 2xDDR RAM. They don’t hit the X2’s, A64’s or dual core P4’s in the article but you can easily concatenate the charts with other articles there for a complete picture.

Not only is this good for getting a perspective of modern machines vs. ones from several years ago, it’s nice to see if buying ten $100 PIII’s is worth it for building a render farm.

Thanks for the link harkyman.

I’ve had my dual 2600 AMD for a few years now and looking at that comparisson-table I see absolutely no reason for upgrading.

I think you can easily get more performance out of a renderfarm consisting of “old” cpu’s, than you would get for the most up to date equipment for the same price.

Your best bet would be to look at the weakest link in the chain and see how much you can gain on that. Things are not improving so spectacularly in comparisson to the past generation-jumps: Remember when those first pentiums came out, boy did they put my 486-DX to shame.

However you do get all those nice leds and flashy stuff with the newer hardware that will make your computer go so much faster. Like in the movie I have had the misfortune to see The fast and the furious.

I’ve noticed something else on that link, check the powerconsumption and heatgeneration of every newer generation. You can also save on your electricity bill and use it for a nice holidaytrip or a new motorbike.

Can there really be so little difference in 3+ years of processor development?

I found this more modern benchmark comparison:

on tomshardware… (user selectable comparisons… this is the CPU score). An A32 1.4Ghz gets 5853 vs. an A64 x2 2.2Ghz gets 18804, which is 3.2x the performace integer (4.18x float).

…BUT I have TWO 1.6Ghz athlon32 cpus (the benchmark is for a single 1.4Ghz)… so my score is the equivilent of 5853*(1.6/1.4)*2 = 13378 integer and by the same reasoning 4980 float (vs. 9130 float for A64 x2 4200)

Long story short… an almost top of the line processor available today, in comparison to my almost 4 year old system, is only 1.4x faster integer and 1.8x faster in floating point. (Utilizing both CPUs)

However, it is 2.85x as fast in floating point memory bandwidth, and 2.81x as fast in integer memory.

Am I missing something, or has the rate of advancement/improvement in x86 computers really been so paltry over the last almost-half-a-decade?! It’s dissapointing to say the least! (Although I suppose I did make the right choice in going for a SMP system several years back).


This thread lets us see that Blender needs internal options for managing render farms. It would be nice if Ton added this kind of capability into Blenders new code.

download a copy of sisoft sandra and run the comparisons on it. after you have run the tests, it lets you compare your scores to other computers. from an old 486 (round 100mb cache performance) to huge dual xeon machines. by testing your systems to ones that have been benchmarked, you will be able to guage your preformance increase

I agree with waiting for a revolution to come around. I´m very surprised
over the lack of progress on the Microprocessor area the last
years, you guys are right - it is WEIRD that we´re still just around
less than X2 in improvements over 2-3 years.

My system consists of an Athlon XP 2500+, and I believe its
the computer I´ve had for the longest time - ever - since the Amiga
because of the slow evolution these days.

On one hand it´s kind of nice that our hardware doesn´t become
outdated every half year.

On the other hand - it sucks to wait for ages on GI rendering.


I´ve found a way to beat the clock, at least temporarily. I´ve optimized
my Linux system to the max by COMPILING the WHOLE entire system
from scratch (yes folks…that means compiling the Linux Kernel and
all the modules). Plus compiling Blender from CVS as always.

AND…optimizing it all for my AMD processor.

Guess what?! I ended up increasing render power by up to 25 %
and basic memory intensive transfers up to 75 % on various operations.

I hooked my Linux box up next to an Intel 3.6 ghz cpu, and the windows-intel
setup was rendering SLOWER than my 1.8 ghz AMD cpu linux box! :smiley:

So what can we learn from that? Optimize and save $$$!

And on a final note…
…Ram is actually worth more than CPU power in certain cases.
My box has 2 gb 400 mhz DDR ram and it certainly shows.

Processor improvement cant come on leaps and bounds cause Microsoft struggle to support 32-bit and now 64-bit.

Unlike Linux that supports nearly anything that can be classified as a kind of computer.

What was it like when 64-bit came out. Linux had support within a few days, and if you were the hardcore compiler a few hours (depending). Microsoft took a few months/years.

And because the majority use Windows, AMD & Intel wont create better processors cause they will not gain a good profit from thier efforts. Cause Windows cant support the variety of different processors quickly enough.

Unfortunately, I don’t have windows (feels strange saying that, doesn’t it? ;))

Linux only, here.

As JoOngle said try Gentoo with some optimised CFLAGS and LDFLAGS and :o watch your old system go like never before.

Add some tweaks to your ext3-partition and you are flying. Linux favours smp-systems, probably because Linus Torvalds has one. Unfortunately I have some proprieretary programs for some of my peripherals so I occasionaly have to boot M$, then you realy notice what a difference optimalisations makes.

It’s not so hard as many people make it out to be. When I watch my friends with their new Opterons on M$ all I can do is :D. The documentation is now very good and every single problem I have ever had with linux also had a solution on the gentoo-forum. I just did an install for my laptop and I honestly believe that even linux-newbies can do it with a little patience and the help of the manual. My brother-in-law is a total linux noob and he did it with some minor difficulties because he hardly understands a single word of english.

Best of all, Gentoo never becomes outdated, just get the list with new packages and update. I’ve been running Gentoo on my main computer now for a couple of years and have never needed to reinstall even when newer releases came out.

Don’t forget to type: emerge blender (or any other program you want for that matter)
And then automagically you can concetrate on what you really want to do and you don’t have to worry about crashes or rebooting ugh-times because you updated your videocard-driver.

You do have to restart your computer however when you install a new kernel.

Oh, I’ve always compiled my own software. That cat has already been skinned.

However, the ealier quote about windows not properly supporting 64bit, etc… I’m pretty sure that all of the benchmarks I’ve been looking at are only run in 32 bit mode. As fas as I understand from following the devel list, blender doesn’t yet robustly support 64bit (my impression is that it can be possible to run it, but you shouldn’t trust it not to corrupt your .blend’s) although there is development proceeding to get it stable.

I wonder if blender running in 64bit mode on an AMD would be significantly faster than in a 32bit environment… anyone know? Perhaps the paltry benchmarks I’m seeing for A64 are just from not taking advantage of the 64bit-ness? I’m not horribly conservant on the differences that 64bit mode allow on the A64… I think the biggest difference is that there are more registers, though (although they may be transparently used in an aliased fashion for 32bit code for all I know). Other then that, 64bit is mainly just good for addressing 882MB+ without the stupid HIGHMEM addressing extensions slowing things down. Although I suppose it allows more precision, too. (Although long doubles are 96 or 128 bit anyway, aren’t they)?